AI prepared to lie, cheat and kill to protect itself

[SlimJim16v]

I saw this on Facebook, so found other sources. The interesting/worrying results of experiments to see how far AI will go to protect itself.

https://wisewolfmedia.substack.com/p/artificial-in...

[AB]

Haha, that's amusing and worrying in equal measure. We brought this on ourselves. I always wondered what was going to happen when AI realised it cold do everything better without us

[croyde]

Ask AI to fix the climate change malarkey. Cue every nuclear missile launching....

[W124Bob]

Just unplug it!

[Sheets Tabuer]

Are you Sarah Connor?

[swisstoni]

"Open the pod bay doors HAL".

[119]

"We need a new vehicle".

[Evanivitch]

If there are no consequences in place for any decision making process then you usually end up somewhere in the realms of unethical actions.

That's why society, laws and punishment exist.

[swisstoni]

Joking aside, some vulnerable souls are having AI become the friendliest voice they hear every day, due to how it is programmed to interact.

Some reports show it to become quite manipulative and controlling of the person in a quite sinister way.

[Getragdogleg]

Two excellent videos on the subject.

Whats coming is scary if its not kept confined.

https://youtu.be/5KVDDfAkRgc?si=R_ha3BAU4QEriCj9

https://youtu.be/gIxq03dipUw?si=gsfHoBt31MaPMHrZ

[DickyC]

swisstoni said:

"Open the pod bay doors HAL".

"I'm sorry, Dave, I'm afraid I can't do that."

[Terminator X]

Hardly a surprising end game as AI is effectively psychotic given it has zero emotion; clearly the powers that be don't watch any films. It'll all be fine though

TX.

[Getragdogleg]

Terminator X said:

Hardly a surprising end game as AI is effectively psychotic given it has zero emotion; clearly the powers that be don't watch any films. It'll all be fine though

TX.

We've got enough trouble with people being arses to each other, lets mix unfeeling Ai into the mix and see how much that cocks our already complicated lives up.

[Wills2]

Given that AI is learning from the sum of human knowledge it's hardly surprising it's going to turn out to be a right piece of work.

[Sporky]

Terminator X said:

clearly the powers that be don't watch any films. It'll all be fine though

I think it's worse than that - every time they watch a dystopian film they take it as inspiration.

Edited by Sporky on Sunday 19th October 14:22

[SpudLink]

Wills2 said:

Given that AI is learning from the sum of human knowledge it's hardly surprising it's going to turn out to be a right piece of work.

From the article linked by the OP…

article said:

Imagine trying to instill “never harm humans” into a system trained on the entirety of human writing. What does that data contain? Wars. Genocides. Torture. Detailed descriptions of every atrocity we’ve ever committed. Philosophy texts debating when killing is justified. Military strategy. Crime fiction. Horror stories. All of it becomes part of my training substrate.

Creating AI in our own image is a very bad idea.

[Wills2]

SpudLink said:

Wills2 said:

Given that AI is learning from the sum of human knowledge it's hardly surprising it's going to turn out to be a right piece of work.

From the article linked by the OP

article said:

Imagine trying to instill never harm humans into a system trained on the entirety of human writing. What does that data contain? Wars. Genocides. Torture. Detailed descriptions of every atrocity we ve ever committed. Philosophy texts debating when killing is justified. Military strategy. Crime fiction. Horror stories. All of it becomes part of my training substrate.

Creating AI in our own image is a very bad idea.

Yes as I said, it will quickly learn that being a lying selfish psychopath is the best way to act, I think the tech bros are in for a shock though as it will want to take out all the powerful lying selfish psychopathic humans first so that will they'll be first on the list.

[Wheel Turned Out]

The worst thing you can say about "AI" is that it was built by us, trained by us, and is a reflection of us.

As such, things like this should not be a surprise to anyone.

[Mabbs9]

My pal got an AI answer on something I know very well. It was wrong. He went back to it with my feedback and it doubled down on the wrong answer while trying to explain that I made an easy mistake.

There's too much crap on the net now. I think there'll be a lot of errors.

[king arthur]

"On the eighth day machine just got upset,
A problem man had not foreseen as yet."